@Article{,
author = {Holyoak, Keith J. and Thagard, Paul},
title = {Analogical mapping by constraint satisfaction},
journal = {Cognitive Science},
year = {1989},
volume = {13},
pages = {295--355}
}
Semantic similarity can be 1) idential relations or 2) sharing common features (e.g. a superordinate category.) [301]
ACME, like SME, takes sets of sentences as input. ACME is not committed to this representation-- other more complex representations could be handled similarly. Propositions are hierarchical, where propositions can take propositions as arguments. [307]
As a whole, the system takes as input: 1) a target description, 2) a source description, 3) semantic weights for certain propositions, and 4) pragmatic information (e.g. proposition D is important.) [309]
Objects can map to objects; propositions to propositions. A unit in the network is created for all possible alignments. n-place predicates can only map to other n-place predicates. In problem solving, propositions in initial states only map to propositions in other initial states, etc.
ACME can make many to one mappings.
Excitatory links are set up, then inhibitory links are put up, discouraging alternative mappings of the same element. [311]
It appears that ACME maps the fortress/tumor problem but does not transfer the solution steps.